diff --git a/src/pkg/sync/atomic/asm_386.s b/src/pkg/sync/atomic/asm_386.s
index 914d2feeb49..99e8b1fd805 100644
--- a/src/pkg/sync/atomic/asm_386.s
+++ b/src/pkg/sync/atomic/asm_386.s
@@ -94,3 +94,9 @@ TEXT ·LoadUint32(SB),7,$0
 	MOVL	0(AX), AX
 	MOVL	AX, ret+4(FP)
 	RET
+
+TEXT ·LoadUintptr(SB),7,$0
+	JMP	·LoadUint32(SB)
+
+TEXT ·LoadPointer(SB),7,$0
+	JMP	·LoadUint32(SB)
diff --git a/src/pkg/sync/atomic/asm_amd64.s b/src/pkg/sync/atomic/asm_amd64.s
index 4282950632a..d21ade1cb60 100644
--- a/src/pkg/sync/atomic/asm_amd64.s
+++ b/src/pkg/sync/atomic/asm_amd64.s
@@ -67,3 +67,11 @@ TEXT ·LoadUint32(SB),7,$0
 	MOVL	AX, ret+8(FP)
 	RET
 
+TEXT ·LoadUintptr(SB),7,$0
+	JMP	·LoadPointer(SB)
+
+TEXT ·LoadPointer(SB),7,$0
+	MOVQ	addrptr+0(FP), AX
+	MOVQ	0(AX), AX
+	MOVQ	AX, ret+8(FP)
+	RET
diff --git a/src/pkg/sync/atomic/asm_linux_arm.s b/src/pkg/sync/atomic/asm_linux_arm.s
index 9ac411944c9..20a45243f26 100644
--- a/src/pkg/sync/atomic/asm_linux_arm.s
+++ b/src/pkg/sync/atomic/asm_linux_arm.s
@@ -96,3 +96,9 @@ loadloop1:
 	BCC	loadloop1
 	MOVW	R1, val+4(FP)
 	RET
+
+TEXT ·LoadUintptr(SB),7,$0
+	B	·LoadUint32(SB)
+
+TEXT ·LoadPointer(SB),7,$0
+	B	·LoadUint32(SB)
diff --git a/src/pkg/sync/atomic/atomic_test.go b/src/pkg/sync/atomic/atomic_test.go
index 2229e58d0c7..5f44bd04b92 100644
--- a/src/pkg/sync/atomic/atomic_test.go
+++ b/src/pkg/sync/atomic/atomic_test.go
@@ -348,6 +348,50 @@ func TestLoadUint32(t *testing.T) {
 	}
 }
 
+func TestLoadUintptr(t *testing.T) {
+	var x struct {
+		before uintptr
+		i      uintptr
+		after  uintptr
+	}
+	var m uint64 = magic64
+	magicptr := uintptr(m)
+	x.before = magicptr
+	x.after = magicptr
+	for delta := uintptr(1); delta+delta > delta; delta += delta {
+		k := LoadUintptr(&x.i)
+		if k != x.i {
+			t.Fatalf("delta=%d i=%d k=%d", delta, x.i, k)
+		}
+		x.i += delta
+	}
+	if x.before != magicptr || x.after != magicptr {
+		t.Fatalf("wrong magic: %#x _ %#x != %#x _ %#x", x.before, x.after, magicptr, magicptr)
+	}
+}
+
+func TestLoadPointer(t *testing.T) {
+	var x struct {
+		before uintptr
+		i      unsafe.Pointer
+		after  uintptr
+	}
+	var m uint64 = magic64
+	magicptr := uintptr(m)
+	x.before = magicptr
+	x.after = magicptr
+	for delta := uintptr(1); delta+delta > delta; delta += delta {
+		k := LoadPointer(&x.i)
+		if k != x.i {
+			t.Fatalf("delta=%d i=%d k=%d", delta, x.i, k)
+		}
+		x.i = unsafe.Pointer(uintptr(x.i) + delta)
+	}
+	if x.before != magicptr || x.after != magicptr {
+		t.Fatalf("wrong magic: %#x _ %#x != %#x _ %#x", x.before, x.after, magicptr, magicptr)
+	}
+}
+
 // Tests of correct behavior, with contention.
 // (Is the function atomic?)
 //
@@ -578,8 +622,8 @@ func TestHammer64(t *testing.T) {
 	}
 }
 
-func hammerLoadInt32(t *testing.T, uval *uint32) {
-	val := (*int32)(unsafe.Pointer(uval))
+func hammerLoadInt32(t *testing.T, valp unsafe.Pointer) {
+	val := (*int32)(valp)
 	for {
 		v := LoadInt32(val)
 		vlo := v & ((1 << 16) - 1)
@@ -597,7 +641,8 @@ func hammerLoadInt32(t *testing.T, uval *uint32) {
 	}
 }
 
-func hammerLoadUint32(t *testing.T, val *uint32) {
+func hammerLoadUint32(t *testing.T, valp unsafe.Pointer) {
+	val := (*uint32)(valp)
 	for {
 		v := LoadUint32(val)
 		vlo := v & ((1 << 16) - 1)
@@ -615,8 +660,40 @@ func hammerLoadUint32(t *testing.T, val *uint32) {
 	}
 }
 
+func hammerLoadUintptr(t *testing.T, valp unsafe.Pointer) {
+	val := (*uintptr)(valp)
+	var test64 uint64 = 1 << 50
+	arch32 := uintptr(test64) == 0
+	for {
+		v := LoadUintptr(val)
+		new := v
+		if arch32 {
+			vlo := v & ((1 << 16) - 1)
+			vhi := v >> 16
+			if vlo != vhi {
+				t.Fatalf("LoadUintptr: %#x != %#x", vlo, vhi)
+			}
+			new = v + 1 + 1<<16
+			if vlo == 1e4 {
+				new = 0
+			}
+		} else {
+			vlo := v & ((1 << 32) - 1)
+			vhi := v >> 32
+			if vlo != vhi {
+				t.Fatalf("LoadUintptr: %#x != %#x", vlo, vhi)
+			}
+			inc := uint64(1 + 1<<32)
+			new = v + uintptr(inc)
+		}
+		if CompareAndSwapUintptr(val, v, new) {
+			break
+		}
+	}
+}
+
 func TestHammerLoad(t *testing.T) {
-	tests := [...]func(*testing.T, *uint32){hammerLoadInt32, hammerLoadUint32}
+	tests := [...]func(*testing.T, unsafe.Pointer){hammerLoadInt32, hammerLoadUint32, hammerLoadUintptr}
 	n := 100000
 	if testing.Short() {
 		n = 10000
@@ -625,11 +702,11 @@ func TestHammerLoad(t *testing.T) {
 	defer runtime.GOMAXPROCS(runtime.GOMAXPROCS(procs))
 	for _, tt := range tests {
 		c := make(chan int)
-		var val uint32
+		var val uint64
 		for p := 0; p < procs; p++ {
 			go func() {
 				for i := 0; i < n; i++ {
-					tt(t, &val)
+					tt(t, unsafe.Pointer(&val))
 				}
 				c <- 1
 			}()
diff --git a/src/pkg/sync/atomic/doc.go b/src/pkg/sync/atomic/doc.go
index b35eb539c05..0f38886601e 100644
--- a/src/pkg/sync/atomic/doc.go
+++ b/src/pkg/sync/atomic/doc.go
@@ -22,6 +22,10 @@
 //
 package atomic
 
+import (
+	"unsafe"
+)
+
 // BUG(rsc): On ARM, the 64-bit functions use instructions unavailable before ARM 11.
 //
 // On x86-32, the 64-bit functions use instructions unavailable before the Pentium.
@@ -62,6 +66,12 @@ func LoadInt32(addr *int32) (val int32)
 // LoadUint32 atomically loads *addr.
 func LoadUint32(addr *uint32) (val uint32)
 
+// LoadUintptr atomically loads *addr.
+func LoadUintptr(addr *uintptr) (val uintptr)
+
+// LoadPointer atomically loads *addr.
+func LoadPointer(addr *unsafe.Pointer) (val unsafe.Pointer)
+
 // Helper for ARM.  Linker will discard on other systems
 func panic64() {
 	panic("sync/atomic: broken 64-bit atomic operations (buggy QEMU)")